A Multi- institutional Prospective Analysis of Impact of CanAssist Breast (Morphometric Immunohistochemistry Based Test) on Adjuvant Chemotherapy Decisions in Early Breast Cancer

Purpose: CanAssist Breast (CAB) has been validated retrospectively for assessing risk of recurrence and thereby usefulness of chemotherapy in HR+/HER2- breast cancer. The objective of this study is to assess the agreement between physician’s treatment plan and CAB risk stratication and evaluate whether CAB results aid in the physician’s treatment decision. Methods: The data on the physician’s treatment plan before and after the CAB test was collected prospectively between 2016 and 2021 in 249 patients. Changes in treatment recommendations and compliance with CAB reports were analyzed. Results: Based on conventional clinicopathological features physicians planned to treat 46% of patients with endocrine therapy (ET) (low-risk-LR)), 24% with chemoendocrine therapy (CET) (high-risk-HR)) and in 30% physicians were uncertain of prescribing chemotherapy (intermediate-risk-IR)) before CAB testing. The correlation between clinical risk assessment and CAB risk stratication (k=0.2 (0.05-0.35) was nonsignicant. CAB classied 64% as LR, which was 18% (9.3-25, P=0.0001) higher compared to clinical LR. In the clinical IR category, CAB risk proportions were 55:45 (LR: HR). We observed a substantial shift in treatment recommendation from CET to ET in 54% (40.75- 66.84, P<0.0001) of clinical HR and ET to CET in 26% (18.27- 35.01, P<0.0001) of clinical LR patients. Overall CAB lead to change in treatment recommendation in 42% of the cohort. Conclusions: There was a signicant impact of CAB on the physician’s treatment decision. CAB provided denite treatment recommendation to IR patients where the physician had dilemma on prescribing chemotherapy and provided precise treatment plan to clinical LR and HR patients.


Introduction:
The inclusion of chemotherapy in the treatment strategy in the ER positive/HER2 negative early-stage breast cancer patients is judged by the risk of cancer relapse. The risk of cancer relapse is assessed by tumor anatomical features like tumor size, node status, histological grade; age of the patient; hormone receptor expression status and Ki-67 expression [1][2][3][4][5][6][7][8][9]. The emerging evidence from different studies indicate that the tumor pathological features do not provide accurate and reliable prognostic information on the disease aggressiveness [10][11]. These features miss on providing the critical information predictive of cancer recurrence. Studies have shown that some node-negative patients require chemotherapy for a better prognosis while some node-positive patients tumors have good prognosis without chemotherapy [12,13]. In fact it has been reported that some patients with small tumors < 2cms bene t of chemotherapy [14,15]. This showcases that prognosis based on clinical factors alone could lead to over or undertreatment in these patients.
Prognostic tests have largely addressed this issue with accurate prognostication with the optimum treatment recommendations [16][17][18][19]. The multi-gene tests with validation data on Caucasian women, might not be appropriate for selection of Asian women who would bene t from chemotherapy due to underlying inherent racial factors beyond clinical parametes of standard of care, contributing to the differences in the prognosis of these patients [20,21]. CanAssist Breast, a prognostic test using immunohistochemistry platform encompassing the crucial 5 biomarkers (CD44, N-Cadherin, pan-Cadherin, ABCC4 and ABCC11) of cancer progression, recurrence and therapy resistant pathways along with clinical parameters (node status, tumor size and histological grade) predicts risk of cancer relapse with the use of a machine learning algorithm [22,23]. The test with considerable validation data on women from India, USA [23] and European countries (under review) has helped a greater number of women from South East countries plan their treatment [24]. The test has shown to have greater than 83% concordance in the low-risk category with the other widely used test, Oncotype DX [25].
Although the test has been in use since 2016, in uence of CAB on physician's advice to offer chemotherapy has not been reported. In this current prospective observational study, we report the impact of CanAssist Breast on individual physician prescribing adjuvant chemotherapy and change in physician's decision about chemotherapy before and after CAB testing in early-stage breast cancer patients.

Methods:
Study population and study design: The study was a prospective, observational, multi-centric involving physicians across the country. The data used for this study was on patients for whom CAB test was prescribed between the period, 2016 to 2021. Before CAB was performed on the tumor samples, the information on the treatment plan for each patient before and after CAB testing was collected through a questionnaire with physician's considerations for giving or avoiding chemotherapy. The physician had no obligation to treat the patient as per CAB test results. The data for the study has been collected on 249 patients from 35 physicians practicing in 30 different hospitals across India. Of these 16 were medical oncologists, 18 were surgical oncologist and one was radiation oncologist. For the study as only the physician's opinion on the treatment plan was obtained from the CAB prescribing physicians as part of their routine clinical practice and did not intervene with the patient's treatment choices, ethics approval or patient consent were not required. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
Sample processing and CAB testing: After the physician referred the patient for CAB testing, the FFPE tumor blocks were shipped to centralised OncoStem laboratory along with histopathology report. Information on tumor characteristics, ER, PR, HER2/neu and Ki-67 staining were as per the histopathology report shared with OncoStem. Tumor content of the FFPE blocks was assessed by haematoxylin and eosin staining. Blocks with ≥ 30% of tumor content were processed for CAB testing. Immunohistochemistry of the ve CAB biomarkers was performed and CAB risk score was obtained using a machine learning algorithm as described earlier [22]. Statistical analysis: The change in percentage of patients who had change in treatment recommendations by CAB and Odd's ratio for estimating the probability of CAB low risk predictions for clinical intermediate and high-risk groups with clinical low-risk group as a reference were analysed by logistic regression using MedCalc.

Results:
Baseline characteristics of patients: The current prospective study consisted of 249 subjects. All were women except two. The median age of diagnosis of the cohort was 59 (26-81 years). Twenty nine percent of the cohort were aged below or equal to 50 years and 71% above 50 years. Thirty one percent of the cohort had patients with T1 tumors, 6% of the patients had tumors greater than 5cm (T3) and 65% had T2 tumors. Median tumor size of the cohort was 2.5cm (range: 0.4-11cm). Seventy nine percent of the cohort had node-negative tumors and 21% with N1 tumors. Sixty seven percent of the patients had tumors with moderately differentiation (G2) and 22% were with poorly differentiated tumors (G3). All patients were with ER positive disease, 94% had PR positive disease. 51% of patients expressed Ki67 greater than 14%. (Table 1).  (Fig. 1a). There was a weak agreement between the two methods of risk assessment with a kappa coe cient of 0.2 (95% CI: 0.05-0.35). CAB assigned 18% more patients to low-risk category (P = 0.0001) (95% CI:9.3 to 25) compared to physician's risk assessment.
Upon restrati cation of clinical risk groups by CAB, we found 26% (n = 30/114) of 'clinical low-risk' were strati ed as high risk by CAB and 54% (n = 33/61) of 'clinical high-risk' as low-risk by CAB. Of the intermediate risk category (n = 74), 55% of the patients were low-risk and 45% high-risk by CAB (Fig. 1b).
In the sub-group analysis, CAB based low-risk proportions were signi cantly (P ≤ 0.05) higher compared to high-risk in all except in N1 subgroup where low-risk and high-risk proportions were 44% and 56% respectively. However, in the sub-groups generated as a result of the physician's risk assessment, the difference in low-and high-risk proportions was nonsigni cant in subgroups with G3 tumors, in patients aged below 50 years and in patients with T3N0 and T1N1 tumors (  Change in the treatment recommendations: The physician's decision of not giving chemotherapy (clinical LR) was majorly based on node-negative status, grade and age of the patient and low Ki67 (Table 3). Similarly, node-positive tumors, grade 3 tumors, age of the patients ≤ 50 years at the time of diagnosis, high ki67 are perceived to be high risk features for prescribing chemotherapy. Patients with a combination of one or more low-risk feature and high-risk features makes physician's decision of prescribing chemotherapy more complicated thereby putting these patients in the 'intermediate risk' category (Table 3).  (Table 4). And in 74 patients (29.7%) in whom the physician was uncertain of prescribing chemotherapy, with CAB test results 41 patients (55%) would be treated with endocrine therapy alone and in 33 (45%) patients' treatment plan would include chemotherapy along with endocrine therapy (Fig. 1b). Assuming that the physicians tend to treat these patients with chemotherapy in the absence of a prognostic test, with 55% (41/74) as low-risk by CAB the overall change in the treatment recommendations was observed in 42% of the cohort (n = 104/249).   (Table 6). risk of recurrence using arti cial intelligence-based approach. The study was taken up with the intention to assess agreement on risk assessment made by the physician and CAB. Here we report the impact of CAB on the physician and the change in his/her treatment plan with CAB test results in a prospective cohort.
As per the physician only 46% of the patients were at low risk for recurrence and could be treated with endocrine therapy alone, whereas with CAB it was 68%. Most importantly CAB could segregate the group of patients for whom the physician was uncertain of treating with chemotherapy, into low and high-risk groups. The physician's choice of treating the patients to give or to withhold chemotherapy largely depended on node status, ki67, grade and age of the patient. In the clinical intermediate risk category, the physicians seem to be perplexed as the patient was associated both with clinical low and high-risk features. It is also noteworthy that in patients with node-negative tumors of size up to 5cm, the physicians seem to be more challenged.  [27]. Similary in N0 sub-group, 13% higher patients were classi ed as HR by CAB compared to clinical HR indicating that CAB would prevent undertreatment in these patients [13].
Other than node status, the size of the tumor equally poses a challenge to physician to decide on Of the clinical HR group there was change from chemo endocrine therapy t o endocrine therapy alone with CAB in 54% (n = 33) of patients, thus preventing overtreatment for these patients. Likewise, CAB advised chemotherapy in 26% (n = 30) of clinical low-risk group wherein otherwise physician would treat these patients with endocrine therapy alone, thereby preventing undertreatment in these patients. Thus overall across all the clinical sub-groups (LR, IR and HR), the change in treatment recommendation (ET to CET; CET to ET) was in 42% (n = 104) of the cohort. CAB which focuses on tumor biological features to determine the aggressiveness of disease provides de nite precise treatment options for the patients.
Some physicians expressed that before treating the patient they would prefer a recommendation from a prognostic test (n = 70) like CanAssist Breast or any other prognostic test. Of these it is noteworthy that 42 patients were categorised as clinical intermediate risk suggesting that physician wants to be guided by a prognostic test. Few other reasons expressed by physician for performing CanAssist Breast were validation of their decision, avoid chemotherapy (Fig. 2).
Physician's adherence to prognostic tests reports is an important aspect. We found that adherance to CAB test results in 94% of patients.
In conclusion these results emphasize the use of a prognostic test for precise treatment decisions in breast cancer patients. The data evidenced the change in chemotherapy treatment plan and how it resolves the con ict of physician in planning the chemotherapy thereby supporting the clinical utility of this comparatively newer test. With the extensive analytical validation and clinical validation data in patients across geographies we believe that CanAssist Breast will be useful in tailoring the chemotherapy decisions in clinical practice.

Declarations:
Funding: The principal investigators of the study received educational grant from OncoStem Diagnostics which is privately funded.

Con icts of interest/competing interests:
Somashekhar S P, MD report of receiving fees for the advisory role from OncoStem Diagnostics. All other authors have no other competing interests to declare.
Code availability:

Not applicable
Availability of data and material: Available